Answers are grounded in this paper and cite their sources.
Identifying standardized field is a fundamental task that facilitates the assessment of standardized field construction progress and precision management. Currently, there is a scarcity of work focused on identifying standardized field due to the challenge posed by the visual similarity between standardized and ordinary field, with differences primarily in the internal distribution of rand. This presents a significant challenge for deep learning models. Change detection in remote sensing has been extensively studied, with one notable method being the SGSLN, an exchanging dual encoder-decoder (EDED) architecture network, renowned for its high recognition accuracy. However, SGSLN has limitations in effectively capturing the intricate spatial dependencies and long-range contextual information present in remote sensing images. To address this, we propose the Trans-SGSLN, a binary change detection model based on semantic guidance and spatial localization, utilizing Gaofen-1 (GF-1) images to identify standardized field. Our novel algorithm transforms dual-temporal high-dimensional features into semantic space and employs a Transformer with non-siamese weights for feature fusion. This enables the network to capture long-range dependencies, effectively handling complex scene changes in remote sensing images. We constructed the standardized grid field transformations (SGFT) dataset using two GF-1 images from the Youyi Farm. The Trans-SGSLN model was trained on this dataset, achieving Precision, Recall, IoU, F1-score, and Accuracy of 83.41%, 69.44%, 61.01%, 75.78%, and 94.47%, respectively. These metrics represent improvements of 3.16%, 0.22%, 1.87%, 1.45%, and 0.43% over the SGSLN model. Comparative analysis on the test set indicates that Trans-SGSLN minimizes false positives and false negatives more effectively than SGSLN, and the algorithm significantly reduces the misidentification of non-standardized field changes. Through segmentation models and parameter experiments, we identified the optimal configuration for Trans-SGSLN, providing a viable solution for large-scale detection of standardized field.
No takes yet. Share an insight, caveat, or question.
Meng et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: